Serissa

有谁在Onet中添加landmark模块吗？

3

在Onet中添加landmark模块，训练时landmark loss不收敛，一直在0.0024附件上下波动，预测的时候也只是四个点预测不准，有做过的大佬可以交流一下吗？

您好！我想对yolov5主干网络采用RM操作然后再采用slimming进行剪枝。我在网络中没有添加mask卷积操作(rmnet_pruning.py中采用)，请问稀疏化训练是要添加在训练yolov5模型的时候还是在RM操作后再进行稀疏化训练，最后进行slimming进行剪枝？

conv输出进入BN后有精度问题？

2

您好！我在做RM的时候发现，conv输出进入BN输出有精度问题？ ```代码部分 idconv1 = nn.Conv2d(self.in_planes, self.in_planes+self.mid_planes, kernel_size=3, stride=1, padding=1, bias=False).eval() idbn1 = nn.BatchNorm2d(self.in_planes+self.mid_planes).eval() # init dirac_ kernel weight, bias, mean var to idconv1 nn.init.dirac_(idconv1.weight.data[:self.in_planes]) bn_var_sqrt1 = torch.sqrt(self.running1.running_var + self.running1.eps) idbn1.weight.data[:self.in_planes]...

请问如何移植到嵌入式设备？

2

请问可以转ncnn支持的caffemodel或者ncnn支持的onnx吗？

我修改了side的值，训练报错。

“C:\Users\Microsoft\Desktop\caffe-linux2Windows\build\ALL_BUILD.vcxproj”(默认目标) (1) -> “C:\Users\Microsoft\Desktop\caffe-linux2Windows\build\python\pycaffe.vcxproj”(默认目标) (23) -> (ClCompile 目标) -> C:\Users\Microsoft\Desktop\caffe-linux2Windows\python\caffe\_caffe.cpp(100): error C2143: 语法错误: 缺少“)”(在“(”的前面) [C:\Us ers\Microsoft\Desktop\caffe-linux2Windows\build\python\pycaffe.vcxproj] C:\Users\Microsoft\Desktop\caffe-linux2Windows\python\caffe\_caffe.cpp(100): error C2206: “InitLogLevelPipe”: typedef 不能用于函数定义 [C:\Users\Microsoft\Desktop\caffe-linux2Windows\build\python\pycaffe.vcxproj] C:\Users\Microsoft\Desktop\caffe-linux2Windows\python\caffe\_caffe.cpp(100): error C2059: 语法错误:“)” [C:\Users\Microsof t\Desktop\caffe-linux2Windows\build\python\pycaffe.vcxproj] C:\Users\Microsoft\Desktop\caffe-linux2Windows\python\caffe\_caffe.cpp(100):...

我在VS2015环境下，测试recognition.cpp函数，执行到net_.reset(new Net(model_file, caffe::TEST));时，提示： I1028 11:15:27.573096 14076 layer_factory.cpp:58] Creating layer data F1028 11:15:27.574095 14076 layer_factory.cpp:62] Check failed: registry.count(type) == 1 (0 vs. 1) Unknown layer type: Input (known types: Convolution, Eltwise,...

Quantitative training official edgeai-yolov5 code does not converge

4

I refer to the official quantitative demo training edgeai-yolov5, but the training did not converge. Quantization training examples/quantization_example.py this demo can converge. Refer to quantization_example.py to modify the code of...

Dear authors, Why is BN polarization regularization used in VGG network and convolution layer polarization regularization used in mobilenetv2? Thank you!